|
|
View previous topic :: View next topic |
Author |
Message |
niq24601
Joined: 07 Jul 2008 Posts: 323
|
|
Back to top |
|
|
ajanata
Joined: 07 Jul 2007 Posts: 1167 Location: South Bay Area, CA
|
Posted: Wed Sep 09, 2009 6:57 pm Post subject: |
|
|
It should be possible, but I'm interested where you go those numbers from. What does your math come up with for just, say, drums, as far as memory usage and time are concerned?
I think you're over-complicating matters. It's just running 4 single-player paths, but taking the unison bonuses and the points from the other parts into consideration. _________________
|
|
Back to top |
|
|
meister
Joined: 09 Nov 2006 Posts: 2257 Location: St Cloud, MN
|
Posted: Wed Sep 09, 2009 7:02 pm Post subject: |
|
|
ajanata wrote: | It should be possible, but I'm interested where you go those numbers from. What does your math come up with for just, say, drums, as far as memory usage and time are concerned?
I think you're over-complicating matters. It's just running 4 single-player paths, but taking the unison bonuses and the points from the other parts into consideration. |
Its not just that, it also has to know where other instruments can activate, which would make it way more complicated _________________
|
|
Back to top |
|
|
Kawigi
Joined: 27 Feb 2008 Posts: 2879 Location: Redmond, WA
|
Posted: Wed Sep 09, 2009 7:39 pm Post subject: |
|
|
meister wrote: | ajanata wrote: | It should be possible, but I'm interested where you go those numbers from. What does your math come up with for just, say, drums, as far as memory usage and time are concerned?
I think you're over-complicating matters. It's just running 4 single-player paths, but taking the unison bonuses and the points from the other parts into consideration. |
Its not just that, it also has to know where other instruments can activate, which would make it way more complicated |
Indeed. If it were as simple as figuring out the optimal paths for each instrument if they were playing with a full band but only they were activating, this would be a really easy and reasonably fast problem. That would work if the multipliers for 1, 2, 3 and 4 people using OD were 2, 3, 4, 5, but they're 2, 4, 6 and 8. This means that in a situation where an instrument could activate by itself in a more point-dense place by itself or activate with others in a place that has at least half as many points, the latter is preferrable (and these restrictions impose themselves because every instrument doesn't get OD at the same time, and some instruments can't activate whenever they want).
Niq's numbers are based on a real algorithm that could work pretty well in theory, but requires a prohibitively huge amount of memory (the proposal he gives is to flush intermediate data to the hard disc, which I suppose can bring the RAM requirements down to the level of reasonability, but notice the staggering amount of temporary storage space we're talking about). I don't know if the time estimate is correct - I suspect it would take somewhat longer, probably a good chunk of a day for one song - but it's hard to really be sure without an implementation that's closer to what he's talking about in scope (and perhaps, running it on Charlene like he suggests). I've done some experimenting with computer-generated two-part paths, which is a pretty closely-related problem that is still (barely) within the realm of fitting in memory with some accuracy (and after some optimizations). _________________
|
|
Back to top |
|
|
niq24601
Joined: 07 Jul 2008 Posts: 323
|
Posted: Thu Sep 10, 2009 1:33 am Post subject: |
|
|
ajanata wrote: | It should be possible, but I'm interested where you go those numbers from. What does your math come up with for just, say, drums, as far as memory usage and time are concerned?
|
For drums it's much easier because of the set number of possible activations. Even if there are activiations at strange places (other than every 4 measures), you end up with something like 128KB. For guitar it's something like 1MB of memory.
As others have pointed out, the problem is that each instrument can activate independently, and there's a reward for activating together, but everyone gets OD at different times ... so the number of possible states for any one half-beat (you need eight-note granularity to avoid a certain amount of rounding) is
(# of instrument combinations in OD) * (# of possible combinations of OD across all instruments)
Since drums and vox can only activate in certain places we estimated this as
16 * 64 * 64 * 32 * 5
Which is about 10 million states *per half-beat*. You need 4 bytes to represent the band states, so that's 40 MB. With twice that space needed for backtracking, and about 2000 half beats in a long song, that's 160 GB. _________________
The Rocktimizer: RB2 Full Band Optimal Paths
"gay" and "rape". You keep on using those words. I do not think they mean what you think they mean. singemfrc wrote: | When I got GHWoR at midnight the guy behind me was there to buy a copy of Iron Man 2 on Blu Ray and midnight. Buying a movie at midnight release? Wow, so I didn't feel like the biggest loser |
|
|
Back to top |
|
|
internetguy87
Joined: 27 Mar 2008 Posts: 3505 Location: Pittsburgh, PA
|
Posted: Thu Sep 10, 2009 1:49 am Post subject: |
|
|
niq24601 wrote: |
Since drums and vox can only activate in certain places we estimated this as
16 * 64 * 64 * 32 * 5
Which is about 10 million states *per half-beat*. You need 4 bytes to represent the band states, so that's 40 MB. With twice that space needed for backtracking, and about 2000 half beats in a long song, that's 160 GB. |
May I just ask where these numbers are exactly from? _________________
|
|
Back to top |
|
|
Kawigi
Joined: 27 Feb 2008 Posts: 2879 Location: Redmond, WA
|
Posted: Fri Sep 11, 2009 3:51 pm Post subject: |
|
|
internetguy87 wrote: | niq24601 wrote: |
Since drums and vox can only activate in certain places we estimated this as
16 * 64 * 64 * 32 * 5
Which is about 10 million states *per half-beat*. You need 4 bytes to represent the band states, so that's 40 MB. With twice that space needed for backtracking, and about 2000 half beats in a long song, that's 160 GB. |
May I just ask where these numbers are exactly from? |
16 is the number of combinations of instruments that could currently have OD active. The 64s (which should actually be 65) represent the amounts of OD that Guitar/Bass will have in half/beats. I can only assume the 32 is vocals OD, but I'm not sure why that number was picked exactly, and the 5 is the amount of OD that drums could have at a given point in the song (sometimes may be more than that, really, but sometimes less, too). _________________
|
|
Back to top |
|
|
niq24601
Joined: 07 Jul 2008 Posts: 323
|
Posted: Fri Sep 11, 2009 11:00 pm Post subject: |
|
|
Kawigi wrote: | I can only assume the 32 is vocals OD, but I'm not sure why that number was picked exactly, and the 5 is the amount of OD that drums could have at a given point in the song (sometimes may be more than that, really, but sometimes less, too). |
That's just a guess that, on average, the vocals will only have half the states reachable in any one beat. I have no idea what the real number will work out to.
And yes, the 64s should have been 65. _________________
The Rocktimizer: RB2 Full Band Optimal Paths
"gay" and "rape". You keep on using those words. I do not think they mean what you think they mean. singemfrc wrote: | When I got GHWoR at midnight the guy behind me was there to buy a copy of Iron Man 2 on Blu Ray and midnight. Buying a movie at midnight release? Wow, so I didn't feel like the biggest loser |
|
|
Back to top |
|
|
|
|
You cannot post new topics in this forum You cannot reply to topics in this forum You cannot edit your posts in this forum You cannot delete your posts in this forum You cannot vote in polls in this forum
|
Copyright © 2006-2024 ScoreHero, LLC
|
Powered by phpBB
|